Adaptive Crystallite Kinetics in Homogenous Bilayer Oxide Memristor for Emulating Diverse Synaptic Plasticity

Abstract

A critical routine for memristors applied to neuromorphic computing is to approximate synaptic dynamic behaviors as closely as possible. A type of homogenous bilayer memristor with a structure of W/HfO y /HfO x /Pt is designed and constructed in this paper. The memristor replicates the structure and oxygen vacancy (V O ) distribution of a complete synapse and its Ca 2+ distribution, respectively, after the forming process. The detailed characterizations of its atomic structure and phase transformation in and near the conductive channel demonstrate that the crystallite kinetics are adaptively coupled with the V O migration prompted by directional external bias. The extrusion (injection) of the V O s and the subsequent crystallite coalescence (separation), phase transformation, and alignment (misalignment) resemble closely the Ca 2+ flux and neurotransmitter dynamics in chemical synapses. Such adaptation and similarity allow the memristor to emulate diverse synaptic plasticity. This study supplies a kinetic process of conductive channel theory for bilayer memristors. In addition, our memristor has very low energy consumption (5-7.5 fJ per switching for a 0.5 µm diameter device, compatible with a synaptic event) and is therefore suitable for large-scale integration used in neuromorphic networks.

Analog–Digital Hybrid Memristive Devices for Image Pattern Recognition with Tunable Learning Accuracy and Speed

“…The left part of Figure d presents the conductance variation of A‐RS. The device conductance is strengthened/weakened continuously by 50 consecutive positive/negative voltage pulses, which can simulate the synaptic potentiation/depression process when treating the device conductance as the synapse weight . The right part of Figure d shows the conductance variation of D‐RS, in which the reversible switching between LCS and HCS is achieved.…”

Bridging Biological and Artificial Neural Networks with Emerging Neuromorphic Devices: Fundamentals, Progress, and Challenges

“…PTP acts on a longer time scale of tens of seconds to minutes. [92,99,118,[176][177][178][179] By varying the time interval between the paired pulses, the amplitude of EPSC caused by the second pulse can be tuned; generally, a longer interval results in a reduced EPSC amplitude enhancement. It is associated with Ca 2+ dependent activation of presynaptic protein kinases.…”

“…The original form of Hebbian learning did not include LTD. CMOS devices, [162] CBRAM, [99] RRAM, [179] hybrid CMOS/RRAM, [224] perovskite devices, [225] and many others have been employed to show SRDP in response to incoming pulses at different frequencies: a high-frequency (20-100 Hz) train of presynaptic pulses results in LTP of the synaptic strength, whereas a low-frequency (1-5 Hz) train results in LTD. [223] SRDP is a terminology used in the field of electronics rather than neuroscience to describe these rules.…”

Ionotronic Neuromorphic Devices for Bionic Neural Network Applications

“…Yin et al reported a bilayer memristor with a W/HfO y /HfO x /Pt structure which replicates the Ca 2+ distribution in biological synapse. Both EPSC and IPSC response were mimicked exhibiting a current relaxation.…”